Using the Fisher-Bingham distribution in stochastic ... - CiteSeerX

18 downloads 814 Views 59KB Size Report
model is the Fisher-Bingham distribution on the 3D sphere (Kent, 1982), .... Physiology, University of Copenhagen, whose support is acknowledged. References.
Using the Fisher-Bingham distribution in stochastic models for protein structure John T. Kent* and Thomas Hamelryck





University of Leeds University of Copenhagen

1 Introduction Protein structure prediction is at present an unsolved problem, partly due to lack of a suitable energy function and partly due to the high number of conformational parameters involved. Many protein structure prediction methods therefore use simplified models (Buchete et al., 2004). A popular approach is to represent a protein by its C  atoms only, ie. by its C trace. Thus a protein can be regarded as a sequence of edges in  . To a first approximation ˚ The angles the length of each edge is constant (in practice one unit of length is about 3.8 A). between successive edges determine the shape of the protein. Here we introduce a probabilistic model of the geometry of a random protein’s C  trace. Such a model has many applications, like for example steering the proposal of protein conformations in a Markov Chain Monte Carlo (MCMC) simulation. The model is based on machine learning techniques, and is trained from a set of representative protein structures. Central to the model is the Fisher-Bingham distribution on the 3D sphere (Kent, 1982), an analogue of the bivariate normal equation in the plane.

2 Simulating the FB  distribution The FB 

  distribution, where and  are real concentration parameters and  is a   orthogonal matrix representing orientation, was introduced in Kent (1982) and defines a statistical model on the unit sphere in   . Its pdf in polar coordinates is given by



  "!$#% &' )(+*-,.0/12,4365789(+*-,7 :;,43>?@,A365BC

where EDGFIH= AJ K denotes the colatitude and coordinates are defined by TU WX

QSR

LDGFMH= NOJP denotes the longitude. TU XW Q V ,A3650Y(Z*[,8 Q R  ,A3650Y,43

Suggest Documents